CytoAtlas

Pan-Disease Single-Cell Cytokine Activity Atlas
Date: February 14, 2026

Executive Summary

CytoAtlas is a comprehensive computational resource that maps cytokine and secreted protein signaling activity across ~29 million human cells and ~31,000 bulk RNA-seq samples from six independent datasets: two bulk RNA-seq resources (GTEx, TCGA) and four single-cell compendia (CIMA, Inflammation Atlas, scAtlas, parse_10M) spanning healthy donors, inflammatory diseases, cancers, and cytokine perturbations. The system uses linear ridge regression against experimentally derived signature matrices to infer activity — producing fully interpretable, conditional z-scores rather than black-box predictions.

29MTotal Cells
31KBulk Samples
6Datasets
1,213Signatures
262API Endpoints
12Web Pages

Key results:

  • 1,213 signatures (43 CytoSig + 1,170 SecAct), plus 178 cell-type-specific LinCytoSig variants, validated across 6 independent datasets
  • Spearman correlations reach ρ=0.6–0.8 at independent levels for well-characterized cytokines (IL1B, TNFA, VEGFA, TGFB family), exceeding ρ=0.9 in specific tissue/cell-type strata
  • Cross-dataset consistency demonstrates signatures generalize across CIMA, Inflammation Atlas Main, scAtlas, GTEx, and TCGA
  • SecAct achieves the highest median correlations in 5 of 6 datasets (independence-corrected median ρ=0.19–0.46)

1. System Architecture and Design Rationale

1.1 Architecture and Processing [Detailed Architecture →]

Linear interpretability over complex models. Ridge regression (L2-regularized linear regression) was chosen deliberately over methods like autoencoders, graph neural networks, or foundation models. The resulting activity z-scores are conditional on the specific genes in the signature matrix, meaning every prediction can be traced to a weighted combination of known gene responses. This is critical for biological interpretation — a scientist can ask “which genes drive the IFNG activity score in this sample?” and get a direct answer.

Reproducibility through separation of concerns. The system is divided into independent components, each chosen for the constraints of HPC/SLURM infrastructure:

ComponentTechnologyPurposeRationale
PipelinePython + CuPy (GPU)Activity inference10–34x speedup over NumPy; batch-streams H5AD files (500K–1M cells/batch) with projection matrix held on GPU; automatic CPU fallback when GPU unavailable
StorageDuckDB (3 databases, 68 tables)Columnar analyticsSingle-file databases require no server — essential on HPC where database servers are unavailable; each database regenerates independently without affecting others
APIFastAPI (262 endpoints)RESTful data accessAsync I/O for concurrent DuckDB queries; automatic OpenAPI documentation; Pydantic request validation; lifespan management for resource initialization
FrontendReact 19 + TypeScriptInteractive exploration (12 pages)Migrated from 25K-line vanilla JS SPA to 11.4K lines (54% reduction) with type safety, component reuse, and lazy-loaded routing

Processing scale. Ridge regression (λ=5×105) is applied using secactpy.ridge() against each signature matrix. For single-cell data, expression is first aggregated to pseudobulk (donor or donor×celltype level), then genes are intersected with the signature matrix (CytoSig: ~4,860 genes; SecAct: ~7,450 genes). The resulting z-scored activity coefficients are compared to target gene expression via Spearman correlation across donors.

DatasetCells/SamplesProcessing TimeHardware
GTEx19,788 bulk samples~10minA100 80GB
TCGA11,069 bulk samples~10minA100 80GB
CIMA6.5M cells~2hA100 80GB
Inflammation Atlas (main/val/ext)6.3M cells~2hA100 80GB
scAtlas Normal2.3M cells~1hA100 80GB
scAtlas Cancer4.1M cells~1hA100 80GB
parse_10M9.7M cells~3hA100 80GB

Total: ~29M single cells + ~31K bulk RNA-seq samples, processed through ridge regression against 3 signature matrices (CytoSig: 43 cytokines, LinCytoSig: 178 cell-type-specific, SecAct: 1,170 secreted proteins). Processing Time = wall-clock time for full activity inference on a single NVIDIA A100 GPU (80 GB VRAM). For bulk datasets (GTEx/TCGA), ridge regression is applied with within-tissue/within-cancer mean centering to remove tissue-level variation. See Section 2.1 for per-dataset details.

1.2 Validation Strategy

CytoAtlas validates at four aggregation levels, each testing whether predicted activity correlates with target gene expression (Spearman ρ) across independent samples:

LevelDescriptionDatasetsReport Section
Donor pseudobulkOne value per donor, averaging across cell typesCIMA, Inflammation Atlas Main, scAtlas Normal/Cancer§4.1, §4.3
Donor × cell-typeStratified by cell type within each donorCIMA, Inflammation Atlas Main, scAtlas Normal/Cancer§4.7
Per-tissue / per-cancerMedian-of-medians across tissues or cancer typesGTEx (29 tissues), TCGA (33 cancer types)§4.2, §4.3
Cross-platformBulk vs pseudobulk concordance per tissue/cancerGTEx vs scAtlas Normal, TCGA vs scAtlas Cancer§4.4
Residual (composition-adjusted)Activity–expression correlation after regressing out cell-type fractionsCIMA, Inflammation Atlas Main, scAtlas Cancer§4.8

All statistics use independence-corrected values — preventing inflation from repeated measures across tissues, cancer types, or cell types. CytoSig vs SecAct comparisons use Mann-Whitney U (total) and Wilcoxon signed-rank (32 matched targets) with BH-FDR correction. See Section 3.3 for the validation philosophy and Section 4 for full results.

Why independence correction matters: Pooling across tissues or cancer types inflates correlations through confounding. For example, GTEx pooled CytoSig median ρ (0.211) is 40% higher than the independence-corrected by-tissue value (0.151); SecAct shows +26% inflation (0.394 vs 0.314). All results in this report use the corrected values. For a detailed comparison of pooled vs independent levels, including inflation magnitude and finer cell-type stratification, see the Section 4.1 statistical supplement.

Figure 1: Dataset Overview
Figure 1. CytoAtlas overview. Data sources (4 single-cell compendia, 2 bulk RNA-seq resources) are processed through ridge regression against 3 signature matrices, then validated across 8 complementary analyses (§4.1–§4.8).

2. Dataset Catalog

2.1 Datasets and Scale [detailed analytics]

#DatasetTypeCells/SamplesDonorsCell TypesReference
1GTExBulk RNA-seq19,788 samples946 donorsGTEx Consortium, v11
2TCGABulk RNA-seq11,069 samples10,274 donorsTCGA PanCancer
3CIMAscRNA-seq6,484,974421 donors27 L2 / 100+ L3J. Yin et al., Science, 2026
4Inflammation Atlas MainscRNA-seq4,918,140817 samples*66+Jimenez-Gracia et al., Nature Medicine, 2026
5Inflammation Atlas ValscRNA-seq849,922144 samples*66+Validation cohort
6Inflammation Atlas ExtscRNA-seq572,87286 samples*66+External cohort
7scAtlas NormalscRNA-seq2,293,951317 donors102 subClusterQ. Shi et al., Nature, 2025
8scAtlas CancerscRNA-seq4,146,975717 donors (601 tumor-only)162 cellType1Q. Shi et al., Nature, 2025
9parse_10MscRNA-seq9,697,97412 donors × 90 cytokines (+PBS control)18 PBMC typesOesinghaus et al., bioRxiv, 2026

Grand total: ~29 million single cells + ~31K bulk samples from 6 independent studies (9 datasets), 100+ cell types.

* Inflammation Atlas does not provide donor-level identifiers; the 817/144/86 values are sample counts. The donor–sample relationship is unknown, so correlations use sampleID as the independent unit.

2.2 Disease and Condition Categories

CIMA (421 healthy donors): Healthy population atlas with paired blood biochemistry (19 markers: ALT, AST, glucose, lipid panel, etc.) and plasma metabolomics (1,549 features). Enables age, BMI, sex, and smoking correlations with cytokine activity.

Inflammation Atlas (19 diseases): RA, SLE, Sjogren's, PSA, Crohn's, UC, COVID-19, Sepsis, HIV, HBV, BRCA, CRC, HNSCC, NPC, COPD, Cirrhosis, MS, Asthma, Atopic Dermatitis

scAtlas Normal (317 donors): 35 organs, 12 tissues with ≥20 donors for per-organ stratification (Breast 124, Lung 97, Colon 65, Heart 52, Liver 43, etc.)

scAtlas Cancer (717 donors, 601 tumor-only): 29 cancer types, 11 with ≥20 tumor-only donors for per-cancer stratification (HCC 88, PAAD 58, CRC 51, ESCA 48, HNSC 39, LUAD 36, NPC 36, KIRC 31, BRCA 30, ICC 29, STAD 27)

parse_10M: 90 cytokines × 12 donors — independent in vitro perturbation dataset for comparison. A considerable portion of cytokines (~58%) are produced in E. coli, with the remainder from insect (Sf21, 12%) and mammalian (CHO, NS0, HEK293, ~30%) expression systems. Because exogenous perturbagens may induce effects differing from endogenously produced cytokines, parse_10M serves as an independent comparison rather than strict ground truth. CytoSig/SecAct has a potential advantage in this regard, as it infers relationships directly from physiologically relevant samples.

2.3 Signature Matrices

MatrixTargetsConstructionReference
CytoSig43 cytokinesMedian log2FC across all experimental bulk RNA-seqJiang et al., Nature Methods, 2021
LinCytoSig178 (45 cell types × 1–13 cytokines)Cell-type-stratified median from CytoSig database (methodology)This work
SecAct1,170 secreted proteinsMedian global Moran's I across 1,000 Visium datasetsRu et al., Nature Methods, 2026 (in press)

3. Scientific Value Proposition

3.1 What Makes CytoAtlas Different from Deep Learning Approaches?

Most single-cell analysis tools use complex models (VAEs, GNNs, transformers) that produce aggregated, non-linear representations difficult to interpret biologically. CytoAtlas takes the opposite approach:

PropertyCytoAtlas (Ridge Regression)Typical DL Approach
ModelLinear (z = Xβ + ε)Non-linear (multi-layer NN)
InterpretabilityEvery gene's contribution is a coefficientFeature importance approximated post-hoc
ConditionalityActivity conditional on specific gene setLatent space mixes all features
ConfidencePermutation-based z-scores with CIOften point estimates only
GeneralizationTested across 6 independent cohortsOften held-out splits of same cohort
BiasTransparent — limited by signature matrix genesHidden in architecture and training data

The key insight: CytoAtlas is not trying to replace DL-based tools. It provides a linear, interpretable readout: when CytoAtlas reports “IFNG activity is elevated in CD8+ T cells from RA patients,” the contributing signature genes and their weights can be examined directly in those cells.

3.2 What Scientific Questions Does CytoAtlas Answer?

  1. Which cytokines are active in which cell types across diseases? — IL1B/TNFA in monocytes/macrophages, IFNG in CD8+ T and NK cells, IL17A in Th17, VEGFA in endothelial/tumor cells, TGFB family in stromal cells — quantified across 19 diseases, 35 organs, and 15 cancer types.
  2. Are cytokine activities consistent across independent cohorts? — Yes. IL1B, TNFA, VEGFA, and TGFB family show consistent positive correlations across all 6 validation datasets (Figure 10).
  3. Does cell-type-specific biology matter for cytokine inference? — For select targets, yes: LinCytoSig improves IL6×Macrophage, VEGFA×Endothelial, and IL2×CD8T prediction in normal tissue, but global CytoSig wins overall and the advantage does not transfer to cancer (Figures 12–13).
  4. Which secreted proteins beyond cytokines show validated activity? — SecAct (1,170 targets) achieves the highest correlations in 5 of 6 datasets (median ρ=0.19–0.46), with novel validated targets like INHBA/Activin A (ρ=0.91 in TCGA Colon), CXCL12 (ρ=0.94 in scAtlas Normal Fibroblast), and BMP family.
  5. Can we predict treatment response from cytokine activity? — We are incorporating cytokine-blocking therapy outcomes from bulk RNA-seq to test whether predicted cytokine activity associates with therapy response. Additionally, Inflammation Atlas responder/non-responder labels enable treatment response prediction using cytokine activity profiles as features.

3.3 Validation Philosophy

CytoAtlas validates against a direct principle: if CytoSig predicts high IFNG activity for a sample, that sample should have high IFNG gene expression. This expression-activity correlation is computed via Spearman rank correlation across donors/samples.

This validation only captures signatures where the target gene itself is expressed. Signatures that act primarily through downstream effectors without upregulating the ligand gene itself would not be captured by this metric.


4. Validation Results

4.1 Overall Performance Summary [Full Details]

PRIMARY independent level: The summary table above reports results at each dataset’s PRIMARY independent level — the aggregation level where samples are fully independent (each donor counted once). This ensures correlation statistics are not inflated by donor duplication. See the “Primary Level” column for each dataset’s level.

How “N Targets” is determined: A target is included in the validation for a given atlas only if (1) the target’s signature genes overlap sufficiently with the atlas gene expression matrix, and (2) the target gene itself is expressed in enough samples to compute a meaningful Spearman correlation. Targets whose gene is absent or not detected in a dataset are excluded. CytoSig defines 43 cytokines and SecAct defines 1,170 secreted proteins. Inflammation Atlas Main retains only 33 of 43 CytoSig targets and 805 of 1,170 SecAct targets because 10 cytokine genes (BDNF, BMP4, CXCL12, GCSF, IFN1, IL13, IL17A, IL36, IL4, WNT3A) are not sufficiently expressed in these blood/PBMC samples.

Stratified levels (GTEx by_tissue, TCGA primary_by_cancer): Correlations are computed within each tissue/cancer type (ensuring independence), then summarized across groups. N Targets counts unique targets at the “all” aggregate level. Finer per-tissue or per-cancer breakdowns are available in Section 4.3 below.

4.2 Cross-Dataset Comparison: CytoSig vs SecAct [Statistical Methods]

Figure 2. Spearman ρ distributions across datasets for CytoSig (43 targets) vs SecAct (1,170 targets). Independence-corrected: GTEx/TCGA use median-of-medians. Mann-Whitney U test p-values shown above each dataset.

Why does SecAct appear to underperform CytoSig in Inflammation Atlas Main?

The total comparison includes ~1,170 SecAct targets vs 43 CytoSig targets, so differences in the non-overlapping targets drive the result. Two complementary tests separate this from matched-target performance:

Total comparison (Mann–Whitney U test): Compares the full ρ distributions of CytoSig (43 cytokine signatures) vs SecAct (~1,170 secreted protein signatures) using independence-corrected values. For GTEx/TCGA, each target’s representative ρ is the median across per-tissue/cancer values (median-of-medians); for other datasets, donor_only/tumor_only ρ is used directly. SecAct achieves a significantly higher median ρ in 5 of 6 datasets (GTEx: p = 4.76 × 10−4; TCGA: p = 2.85 × 10−3; CIMA: p = 3.18 × 10−2; scAtlas Normal: p = 1.04 × 10−4; scAtlas Cancer: p = 1.06 × 10−5). Inflammation Atlas Main is the sole exception (U = 14,101, p = 0.548, not significant) and the only dataset where CytoSig’s median ρ (0.323) exceeds SecAct’s (0.173).

Matched comparison (Wilcoxon signed-rank test): Restricts to the 32 targets shared between both methods (22 direct + 10 alias-resolved), each target serving as its own control. SecAct’s median ρ is consistently higher across all 6 datasets, reaching significance in 5 (GTEx: p = 3.54 × 10−5; TCGA: p = 3.24 × 10−6; CIMA: p = 2.28 × 10−2; scAtlas Normal: p = 3.54 × 10−5; scAtlas Cancer: p = 3.54 × 10−5). Inflammation Atlas Main is not significant (p = 0.141).

Inflammation Atlas Main is largely blood-derived; 99 SecAct targets show negative ρ only in this dataset while remaining positive in all other datasets. The “Matched” tab above restricts the comparison to the 32 shared targets, removing the effect of non-overlapping targets.

4.3 Per-Tissue and Per-Cancer Stratified Validation [Statistical Methods]

Figure 3. Per-tissue/cancer CytoSig vs SecAct median Spearman ρ comparison. BH-FDR corrected significance: *** q<0.001, ** q<0.01, * q<0.05.

Stratified validation: Instead of aggregating tissues/cancers into a single median-of-medians, this view shows the CytoSig vs SecAct comparison within each individual tissue (GTEx) or cancer type (TCGA). Mann-Whitney U test (Total tab: all targets) and Wilcoxon signed-rank test (Matched tab: 32 shared targets) with BH-FDR correction across all strata within each dataset.

Key insight: On matched targets (23 pairs per GTEx tissue, 19–21 per TCGA cancer after expression filtering), SecAct wins direction in 29/29 GTEx tissues and 32/33 TCGA cancer types, with 23/29 and 31/33 reaching significance (q<0.05). The sole exception is Acute Myeloid Leukemia (Δ=−0.08, q=0.83). This consistency across 61 of 62 strata indicates the SecAct advantage holds at the individual tissue/cancer level, not only in aggregate. On total targets, SecAct wins in 28/29 GTEx tissues (21 significant) and 30/33 TCGA cancers (15 significant); the few CytoSig-favored strata (Brain in GTEx; Kidney Chromophobe, Ovarian, Uveal Melanoma in TCGA) are all non-significant. Because this comparison uses the same 23–32 matched cytokines, the difference reflects per-target signature performance rather than target count. The largest Δ values occur in Small Intestine (+0.55), Vagina (+0.34), Pancreatic (+0.34), and Head & Neck (+0.27); the smallest in Heart (+0.06), Skin (+0.06), Uveal Melanoma (+0.07), and Kidney Chromophobe (+0.10).

4.4 Cross-Platform Comparison: Bulk vs Pseudobulk [Statistical Methods]

Figure 4. Cross-platform concordance: per-target Spearman ρ distributions from bulk RNA-seq (GTEx/TCGA) vs single-cell pseudobulk (scAtlas) for matching tissues/cancer types. Side-by-side boxplots show the correlation distribution for each tissue/cancer.

Cross-platform concordance: This section tests whether expression–activity relationships replicate across measurement technologies. For each tissue (GTEx) or cancer type (TCGA), we compute per-target Spearman ρ from bulk RNA-seq data and compare it to the same target’s ρ from single-cell pseudobulk data (scAtlas). Wilcoxon signed-rank tests (paired by target) with BH-FDR correction assess whether ρ values differ between platforms.

Key finding: Using all targets, SecAct shows significant bulk–pseudobulk differences in most strata (11/13 GTEx tissues, 5/11 TCGA cancers), while CytoSig shows almost none (1/13, 0/11). When restricted to the same 32 shared targets, both CytoSig and SecAct show no significant platform differences (0/13 and 0/13 for GTEx; 0/11 and 1/11 for TCGA). Matched and unmatched SecAct targets show the same per-target platform shift (mean |Δ| = 0.298 vs 0.302, Mann–Whitney p = 0.82), but SecAct’s ~1,000 paired targets per tissue provide 25× more observations than CytoSig’s ~40, allowing detection of a small systematic shift (Δ ≈ 0.03) that is not detectable with CytoSig’s sample size.

4.5 Best and Worst Correlated Targets

Figure 5. Top 15 (best) and bottom 15 (worst) correlated targets. Select signature type and dataset from dropdowns.

All 32 matched targets have data in ≥4 datasets. Using donor-level aggregation (matching Figure 5), two categories emerge (mean ρ across datasets):

Consistent in Both CytoSig and SecAct (16 targets, both mean ρ > 0.25)
TargetCytoSig Mean ρRangeSecAct Mean ρDatasets ρ>0.2
IL1B+0.56+0.24 to +0.72+0.586/6 & 6/6
TNFA+0.50+0.26 to +0.63+0.466/6 & 6/6
IFNG+0.44+0.30 to +0.62+0.356/6 & 5/6
IL1A+0.43+0.03 to +0.71+0.335/6 & 5/6
IL27+0.43+0.19 to +0.56+0.405/6 & 5/6
TGFB3+0.39+0.19 to +0.53+0.425/6 & 5/6
IL6+0.38+0.17 to +0.53+0.485/6 & 6/6
OSM+0.38+0.06 to +0.49+0.575/6 & 6/6
LIF+0.37+0.20 to +0.62+0.505/6 & 6/6
IL10+0.35+0.06 to +0.55+0.565/6 & 5/6
CXCL12+0.34+0.10 to +0.57+0.594/6 & 5/6
TGFB1+0.34+0.05 to +0.56+0.414/6 & 5/6
BMP4+0.33−0.02 to +0.61+0.434/6 & 5/6
BMP2+0.31+0.19 to +0.41+0.455/6 & 6/6
Activin A+0.29+0.12 to +0.54+0.564/6 & 6/6
GMCSF+0.26+0.01 to +0.46+0.374/6 & 4/6
SecAct-Only: CytoSig Near-Zero, SecAct Rescues (11 targets)
TargetCytoSig Mean ρRangeSecAct Mean ρΔ
LTA−0.02−0.33 to +0.26+0.53+0.55
HGF+0.06−0.29 to +0.40+0.58+0.51
TWEAK−0.02−0.22 to +0.11+0.44+0.47
IL15+0.11−0.05 to +0.43+0.57+0.47
BMP6+0.04−0.40 to +0.26+0.49+0.45
TRAIL−0.00−0.54 to +0.58+0.44+0.44
CD40L+0.02−0.55 to +0.57+0.46+0.43
FGF2+0.04−0.23 to +0.29+0.46+0.42
MCSF+0.16−0.26 to +0.50+0.40+0.24
IL21−0.02−0.22 to +0.09+0.22+0.24
BDNF+0.11−0.07 to +0.20+0.33+0.21

Remaining 5 targets do not fit either category: VEGFA and IFNL are CytoSig-only (CytoSig mean +0.38/+0.21, SecAct < 0.2), GDF11 and GCSF are borderline (CytoSig mean +0.23 each, SecAct +0.43/+0.35), and IL36 shows near-zero correlations in both methods.

Key insight: 16 of 32 matched targets show mean ρ > 0.25 for both CytoSig and SecAct. For 11 additional targets, CytoSig averages near zero while SecAct achieves mean ρ > 0.2 (e.g., TWEAK: CytoSig −0.02 vs SecAct +0.44; LTA: −0.02 vs +0.53; HGF: +0.06 vs +0.58). Note: this section uses donor-level aggregation (matching Figure 5), not the independence-corrected median-of-medians used in §4.1–4.3. Select “SecAct” in the dropdown above to explore interactively.

4.6 Cross-Atlas Consistency

Figure 6. Key cytokine target correlations tracked across 6 independent datasets (independence-corrected levels). Solid lines = CytoSig; dashed lines = SecAct. Lines colored by cytokine family. Click legend entries to show/hide targets.
TierTargetsPattern
Universal (ρ>0.1 in all 6)IL1B, TNFA, IFNG, IL6, BMP2, VEGFARobust across all cohorts and platforms (mean ρ 0.31–0.58)
Mostly consistent (4–5 of 6)IL10, TGFB1, CXCL12, GMCSF, HGFOccasional near-zero outliers; TGFB1 slightly negative only in scAtlas Normal (−0.05)
Context-dependent (≤3 of 6)IL4, IL17A, EGFSign changes across cohorts; Th2/Th17 cytokines absent from Inflammation Main

Key insight: Mean |ρ| across 14 targets does not separate by platform: GTEx 0.26, TCGA 0.33, CIMA 0.26, Inflammation Main 0.44, scAtlas Normal 0.36, scAtlas Cancer 0.37. The universal tier (IL1B, TNFA, IFNG, IL6) shows ρ > 0.1 in all 6 datasets; context-dependent targets (IL4, IL17A) show sign changes and are absent from Inflammation Main.

4.7 Effect of Aggregation Level [Statistical Methods]

Figure 7. Effect of cell-type annotation granularity on validation correlations. Total: CytoSig (43 targets) vs SecAct (1,170 targets). Matched: 32 shared targets only. Select atlas from dropdown.

Aggregation levels explained: Pseudobulk profiles are aggregated at increasingly fine cell-type resolution. At coarser levels, each pseudobulk profile averages more cells, yielding smoother expression estimates but masking cell-type-specific signals. At finer levels, each profile is more cell-type-specific but based on fewer cells.

Key insight: Beyond ~L2 annotation depth, % positive correlations approach 50% and signal/null ratio drops below 2×. Inflammation Main retains signal at L2 (553 samples/group) while scAtlas Normal shows weaker correlations at its shallowest stratification (22 samples/group). SecAct’s larger target pool (~1,170) shows higher retention than CytoSig (43 targets) in aggregate, but on matched targets the difference disappears.

AtlasLevelDescriptionN Cell Types
CIMA Donor OnlyWhole-sample pseudobulk per donor1 (all)
Donor × L1Broad lineages (B, CD4_T, CD8_T, Myeloid, NK, etc.)7
Donor × L2Intermediate (CD4_memory, CD8_naive, DC, Mono, etc.)28
Donor × L3Fine-grained (CD4_Tcm, cMono, Switched_Bm, etc.)39
Donor × L4Finest marker-annotated (CD4_Th17-like_RORC, cMono_IL1B, etc.)73
Inflammation Atlas Main Donor OnlyWhole-sample pseudobulk per donor1 (all)
Donor × L1Broad categories (B, DC, Mono, T_CD4/CD8 subsets, etc.)18
Donor × L2Fine-grained (Th1, Th2, Tregs, NK_adaptive, etc.)65
scAtlas Normal Donor × OrganPer-organ pseudobulk (Bladder, Blood, Breast, Lung, etc.)25 organs
Donor × Organ × CT1Broad cell types within each organ191
Donor × Organ × CT2Fine cell types within each organ356
scAtlas Cancer All Tumor CellsWhole-sample pseudobulk per tumor donor1 (all)
Per Cancer TypePseudobulk stratified by cancer type (HCC, PAAD, CRC, etc.)29 types
Per Cancer Type × CT1Broad cell types within each cancer type~120

4.8 Residual Correlation: Cell-Fraction Adjustment

Direct correlations between predicted activity and target gene expression may be confounded by cell-type composition: if a donor has many monocytes, both TNF expression and predicted TNF activity will be high—not because the prediction is accurate, but because both are driven by monocyte abundance. This section removes that confound.

Method: For each target, we fit X = b0 + b1·F + b2·(A×F) + ε, where F = cell-type fractions and A×F = interaction terms. The residual correlation Spearman(ε, A) tests whether predicted activity associates with expression independent of cell-type composition.

Applied to 3 single-cell datasets with matched cell-fraction data: CIMA (6 L1 cell types), Inflammation Atlas Main (15 L1 cell types), and scAtlas Cancer (15 cell types incl. tumor). scAtlas Normal is excluded due to incompatible donor identifiers between pseudobulk files.

Figure 8. Dumbbell chart: each target is a row. Blue circles = direct ρ (unadjusted), red diamonds = residual ρ (composition-adjusted). Connecting lines show the shift: gray = minimal change (<0.05), orange = decreased, green = increased. Targets sorted by direct ρ descending.

Key finding: In PBMC-dominated datasets (CIMA, Inflammation Main), most direct correlations collapse after cell-fraction adjustment: CIMA retains only 29% of positive correlations (CytoSig) and Inflammation Main retains 43%, with median Δρ of −0.16 and −0.30 respectively. This suggests that donor-level correlations in these datasets are largely driven by cell-type composition rather than within-cell-type biology.

In contrast, scAtlas Cancer retains 97% of positive CytoSig correlations (median Δρ=−0.09) and 88% for SecAct (Δρ=−0.17). This is expected: tumor cells dominate the cell composition, so inter-donor variation in tumor cell fraction is smaller relative to within-tumor biological variation.

4.9 Representative Scatter Plots

Figure 9. Donor-level expression vs predicted activity. Select target, atlas, and signature method from dropdowns.

Key patterns in scatter data: IL1B is the most consistently well-correlated target across all 6 datasets and both signatures (ρ=0.37–0.77). CD40L and TRAIL show a CytoSig-specific sign reversal in CIMA (ρ=−0.48/−0.46) and Inflammation Main (ρ=−0.55/−0.54), while SecAct maintains positive correlations for the same targets (CD40LG ρ=0.14–0.71; TNFSF10 ρ=0.31–0.77 except scAtlas Normal −0.004).

HGF shows a large method difference: CytoSig yields negative or weak correlations in CIMA (−0.25), Inflammation Main (−0.30), and TCGA (0.20), while SecAct achieves 0.64, 0.36, and 0.64 respectively. VEGFA is the most dataset-dependent target—the highest CytoSig ρ in Inflammation Main (0.79) but weak with SecAct in GTEx (0.10) and CIMA (−0.24). Bulk datasets (GTEx n=19,788; TCGA n=11,069) show consistently positive correlations across nearly all targets, while smaller cohorts (CIMA n=421; Inflammation Main n=817) show larger between-target variance and more sign reversals.

4.10 Biologically Important Targets Heatmap

Figure 10. Spearman ρ heatmap for biologically important targets across all datasets. Switch between signature types. Hover over cells for details.

How each correlation value is computed: For each (target, atlas) cell, we compute Spearman rank correlation between predicted cytokine activity (ridge regression z-score) and target gene expression across all donor-level pseudobulk samples. Specifically:

  1. Pseudobulk aggregation: For each atlas, gene expression is aggregated to the donor level (one profile per donor or donor × cell type).
  2. Activity inference: Ridge regression (secactpy.ridge, λ=5×105) is applied using the signature matrix (CytoSig: 4,881 genes × 43 cytokines; SecAct: 7,919 genes × 1,170 targets) to predict activity z-scores for each pseudobulk sample.
  3. Correlation: Spearman ρ is computed between the predicted activity z-score and the original expression of the target gene across all donor-level samples within that atlas. A positive ρ means higher predicted activity tracks with higher target gene expression.

GTEx uses per-tissue pseudobulk (median-of-medians across 29 tissues); TCGA uses per-cancer type (median-of-medians across 33 cancers); CIMA/Inflammation Atlas Main use donor-only; scAtlas Normal uses donor-only; scAtlas Cancer uses tumor-only.

Key insight: A core set of ~10 CytoSig targets shows positive ρ (>0.15) across all 6 datasets (IL1B, TNFA, IL6, IL1A, IL27, IFNG, TGFB3, LIF, BMP2, VEGFA). Conversely, IL2, IL22, IL21, IL3, and IL17A show near-zero or negative ρ across all datasets. SecAct outperforms CytoSig on 26 of 32 matched targets, with the largest differences for LTA (CytoSig ρ≈0.01 vs SecAct ρ≈0.46), CD40L (0.06 vs 0.50), and IL15 (0.03 vs 0.56). CD40L, HGF, and TRAIL flip sign between datasets (e.g., HGF is +0.65 in scAtlas Normal but −0.30 in Inflammation Main).

4.11 Per-Target Correlation Rankings

Figure 11. Validation: targets ranked by Spearman ρ across all datasets and signature types. Select dataset and signature from dropdowns. For SecAct, green bars indicate the 32 matched CytoSig targets (22 direct + 10 alias-resolved); gray bars are additional top-ranked SecAct targets.

Key insight: CytoSig target rankings are consistent across datasets: IL1B, IL10, TGFB3, and IL1A appear in the top 10 of 5/6 datasets, while IL2, IL22, and IL3 appear in the bottom 10 of 5/6 datasets. SecAct rankings are compressed into a narrower, higher range: in scAtlas Normal, every displayed SecAct target exceeds ρ=0.21, whereas CytoSig ranges from −0.42 to +0.69. Inflammation Main has the widest CytoSig range (ρ from −0.55 to +0.79) with only 33 targets passing the filter; VEGFA (ρ=0.79) and IL1A (ρ=0.69) reach their highest values in this dataset.


5. CytoSig vs LinCytoSig vs SecAct Comparison

[Open Issues & Analysis →]    [Full 10-Way Method Comparison →]

Can cell-type-specific signatures outperform global CytoSig for specific targets? We test 7 biologically motivated celltype–cytokine pairs (Macrophage×IL6, HUVEC×VEGFA, T Cell×IL2, Macrophage×IFNG, Macrophage×IL10, Macrophage×TNFA, Fibroblast×TGFB1) across all 6 datasets, comparing CytoSig, LinCytoSig, and SecAct.

5.1 Donor-Level: 7 Representative Targets

Figure 12. CytoSig vs LinCytoSig vs SecAct for 7 representative celltype–cytokine pairs at donor level. “All Datasets” shows per-dataset median ρ across 7 targets; individual dataset views show per-target ρ values. LinCytoSig uses the biologically matched cell-type signature (e.g., Macrophage__IL6 for IL6).

Key findings (donor-level, 7 targets):

  • LinCytoSig wins the median in 2 of 6 datasets (Inflammation Atlas Main, scAtlas Normal) — datasets with rich immune cell diversity.
  • Per-target: LinCytoSig wins 15/42 comparisons vs CytoSig’s 11/42 (SecAct: 16/42). IL2 and IFNG are LinCytoSig’s strongest targets; Fibroblast__TGFB1 is the main drag.
  • SecAct wins the median in 4 of 6 datasets (GTEx, TCGA, CIMA, scAtlas Cancer), driven by its broad coverage of secreted proteins.

Full per-target analysis →

5.2 Celltype-Level Evaluation

Figure 13. CytoSig vs LinCytoSig vs SecAct evaluated on matched cell-type pseudobulk (e.g., IL6 on macrophage pseudobulk) in scAtlas Normal and scAtlas Cancer. Only these two datasets have sufficient cell-type diversity for all 7 targets.

Key findings (celltype-level):

  • LinCytoSig excels where biology matches: IL6×Macrophage, VEGFA×Endothelial, and IL2×CD8T in scAtlas Normal — cytokines with strong cell-type specificity to the tested cell type.
  • Advantage does NOT transfer to cancer: LinCytoSig loses for IL6 and VEGFA in scAtlas Cancer, likely because tumor-associated cells have different transcriptional programs.
  • CytoSig wins TNFA and TGFB1 consistently across all datasets at celltype level — the global signature captures these signaling responses better even on the “right” cell type.

CIMA and Inflammation Atlas Main lack macrophage, endothelial, and fibroblast annotations — only 2–3 of 7 targets evaluable using Myeloid/Mono proxies. Full celltype analysis →

5.3 Limitations: Experimental Bias in the CytoSig Database

The CytoSig database has systematic experimental bias: it preferentially acquires specific cell types (macrophage, fibroblast, cancer lines) and specific cytokines (IFNG, TGFB1, IL6). Cell types with fewer than 10 experiments per cytokine produce noisy signatures, making it difficult to systematically select the “best” cell-type-specific signature for each target.

The cell-type-specific approach has merit for well-characterized targets (IL6×Macrophage, VEGFA×Endothelial, IFNG×Macrophage) but the current CytoSig database cannot support systematic evaluation. Future work could use perturbation data (parse_10M: 90 cytokines × 12 donors × 18 cell types) or expanded cell-type databases to revisit this question.

Full Issues & Analysis →


6. Key Takeaways for Scientific Discovery

6.1 What CytoAtlas Enables

  1. Quantitative cytokine activity per cell type per disease — 43 CytoSig cytokines + 1,170 SecAct secreted proteins across 29M cells
  2. Cross-disease comparison — same signatures validated across 19 diseases, 35 organs, 15 cancer types
  3. Independent perturbation comparison — parse_10M provides 90 cytokine perturbations × 12 donors × 18 cell types for independent comparison with CytoSig predictions
  4. Multi-level validation — donor, donor × celltype, bulk RNA-seq (GTEx/TCGA), and resampled bootstrap validation across 6 datasets

6.2 Limitations

  1. Linear model: Cannot capture non-linear cytokine interactions
  2. Transcriptomics-only: Post-translational regulation invisible
  3. Signature matrix bias: Underrepresented cell types have weaker signatures
  4. Validation metric: Expression-activity correlation underestimates true accuracy (signatures acting through downstream effectors are not captured)

6.3 Future Directions

  1. scGPT cohort integration (~35M cells)
  2. cellxgene Census integration
  3. Classification of cytokine blocking therapy

7. Appendix: Technical Specifications

A. Computational Infrastructure

B. Statistical Methods

C. Detailed System Architecture

A comprehensive technical reference covering both cytoatlas-pipeline (offline GPU computation) and cytoatlas-api (online serving) is available as a separate document.

View Detailed System Architecture →